You are on page 1of 22

CHAPTER 3

Radiative transfer modeling in


the cryosphere
Roberto Furfaro, Alberto Previti, Paolo Picca, Jeffrey S. Kargel, and Michael P. Bishop

ABSTRACT ical to obtaining meaningful and accurate RT


calculations. The common methods employed to
Radiative transfer (RT) modeling plays a key role in determine single-scattering albedo and scattering
interpreting the radiance measured by multispectral phase function, for both single-type particles and
sensors. Glaciers respond to variations in solar irra- mixtures, are discussed. In addition, although the
diance. At-sensor radiance depends upon glacier basic conservation of photons holds for both
surface material composition and intermixture of glaciers and glacier lake water, we have marked a
materials, solar and sensor geometry, and surface clear distinction between the equation of transfer
topography. To bridge the gap between investiga- for glacier surfaces and glacier lake water, as well as
tive findings and spectral data, a physically based between the methods employed to describe their
(i.e., based on first principles) linkage between optical properties. The chapter also provides
properties of the observed surface and the measured examples of RT-based calculations for both BRF
electromagnetic signal should be established. Com- and spectral albedo in scenarios typically found in
plementing the treatment of related subjects in the cryosphere. Five simulation sets show how
Chapter 2 of this book by Bishop et al., in this remotely measurable quantities depend on the mor-
chapter we show how RT theory can be adapted phological and mineralogical properties of the med-
to derive radiative transfer equations (RTEs) that ium (e.g., BRF for mixtures of snow and debris;
are commonly employed to properly describe the spectral albedo variation for snow and carbon soot
radiative field within, at the surface of, and above with varying grain size and particle concentration;
glaciers, debris fields, and glacier lakes. RTEs are and spectral variation of glacier lake water reflec-
derived using the basic principle of conservation of tance as a function of rock ‘‘flour’’ concentration).
photons and are simplified to obtain equations that
are more mathematically tractable. Such equations
are numerically solved to compute quantities that 3.1 INTRODUCTION
are of interest in remote sensing (e.g., bidirectional
reflectance factor, BRF, and spectral albedo) that Remote sensing of the Earth’s cryosphere is an
are a function of the optical properties of the active research area, as glaciological processes are
observed surface. Accurate modeling of the optical closely linked to atmospheric, hydrospheric, and
properties of single-material particles (e.g., ice or lithospheric processes (Bush 2000, Shroder and
snow, water, lithic debris, and carbon soot) is crit- Bishop 2000, Meier and Wahr 2002) and a host
54 Radiative transfer modeling in the cryosphere

of issues of practical human concern. Global under- face morphology, and composition variations.
standing of cryospheric processes involves analysis Glacier surfaces are generally comprised of a
of glacier dynamics since they are affected by and variety of materials and exhibit a complex reflec-
can influence climate change (Kotlyakov et al. 1991, tance distribution depending upon the spatial struc-
Seltzer 1993, Haeberli and Beniston 1998, Maisch ture of surface constituents. Spatial and temporal
2000). Thus, characterization and estimation of variations in debris cover and intimate or areal
glacier surface properties, such as ice grain size, mixtures between coarse-grained glacier ice, snow,
rock debris cover, and surface water distribution, liquid water, vegetation, and rock debris contribute
become critical to advancing our understanding of to highly variable reflectance as observed by in situ
glacier–climate relationships and glacier fluctua- and platform-based sensors (Kargel et al. 2005).
tions (Bishop et al. 2004, Kargel et al. 2005). Modeling plays a central role in investigating the
Because of the ability of orbiting platforms to relationships between surface mixtures and reflec-
provide global and continuous coverage of vast tance, and can assist glacier mapping and charac-
portions of Earth’s surface, satellite imagery can terization. BRDF and BRF modeling are important
be processed to extract information about impor- functional components of scientific inquiry because
tant surface properties. Most space-based sensors they help bridge the gap between investigative
operate passively, measuring the magnitude of findings and field-based and remote observations.
reflected/emitted surface radiance in the visible, For example, Mishchenko et al. (1999) modeled the
infrared, and thermal portions of the spectrum. directional reflectance pattern and its effect on
The Advanced Spaceborne Thermal Emission and albedo for four types of soils, each characterized
reflection Radiometer (ASTER), Moderate Resolu- by a different index of refraction. The BRF patterns
tion Imaging Spectrometer (MODIS), Polarization were generated for snow using three different
and Directionality of Earth Reflectance instrument scattering phase functions (hexagonal ice, fractal
(POLDER), and Multi-angle Imaging Spectro- ice, and spherical ice) to examine the effect of ice
Radiometer (MISR) are examples of current morphology on reflected radiation.
space-based instruments that collect such multi- Models can also be used to provide a basis for
spectral data. Each of these instruments can record surface parameter retrieval. For example, Piatek et
the directionality and intensity of surface radiance, al. (2004) use model inversion and laboratory-based
thereby recording information about scattering data to extract the basic optical properties of rego-
properties that are a function of wavelength. For lith as a function of the chemical composition and
icy surfaces on Earth, we require spatial and tem- grain size. Radiative transfer modeling was also
poral information about: (1) the amount of clear used by Painter et al. (2003) to generate a lookup
and dirty ice; (2) snow coverage and rock debris table comprising BRF patterns for spherical ice
(including both patchy and intimate mixtures); with variable grain size. Each of the elements of
(3) the distribution of ponded or flowing water; the lookup table was considered an endmember
(4) the liquid water content of snow and grain size of the ice family, and a linear unmixing algorithm
of snow and ice; and (5) the extent of partial vegeta- was used to map grain-size distribution over alpine
tion cover (Kargel et al. 2005). These compositional glaciers.
and phase-state parameters are critical to character- Radiative transfer (RT) theory is the logical
izing the current state and recent dynamics of quantitative framework and physical basis for mod-
glaciers and, consequently, they are key to obtain- eling spectral and directional reflectance as a func-
ing a better understanding of the processes asso- tion of surface composition and granular texture.
ciated with the cryospheric impacts of climate Its theoretical basis was established by the seminal
change. work of Chandrasekhar (1960), who derived the
Quantitative characterization of surface basic radiative transfer equation (RTE) describing
properties using remotely sensed data requires the the transport of photons moving in a generic med-
definition of a functional relationship between ium characterized by specified optical properties.
surface properties and surface radiance. The bi- He also derived a set of techniques to determine
directional reflectance distribution function analytical and numerical solutions for a large
(BRDF) and/or the closely related bidirectional variety of radiative transfer problems. Whereas
reflectance factor (BRF), as well as the spectral the RTE is capable of describing the radiative
albedo, are the typical parameters employed to regime within a prescribed host medium, most of
describe surface reflectance given irradiance, sur- the efforts involving solutions of the RTE focused
Radiative transfer modeling of glacier surfaces 55

on determining the number of photons reflected by rock debris) and, separately, the optical properties
a planetary surface as a function of incident radia- of glacier water. Methods currently employed to
tion, viewing geometry, and surface properties. The model the optical properties of mixtures are also
goal was to provide a means for quantitative inter- reported. Section 3.4 briefly reviews the variety of
pretation of the signal collected by airborne and/or numerical methods available to solve the RTE and
spaceborne instruments. For example, Hapke discusses some of the RT codes available in the
(1981, 1986, 1993, 2002, 2008) devised a semi- literature. We then show some examples of simula-
empirical RTE-based bidirectional reflectance tions of albedo and BRF for a variety of scenarios
model that, under simplifying assumptions, allowed that are typically found on glacier surfaces.
the analytical computation of the surface reflec-
tance factor and other photometric properties of
interest. The Hapke model has been successful in
modeling the reflectance factor of a variety of
planetary surfaces in the solar system (e.g., Helfen- 3.2 RADIATIVE TRANSFER MODELING
stein et al. 1988, Warell et al. 2009) and it is very OF GLACIER SURFACES
popular in the planetary science community.
More recently, Mishchenko et al. (1999) devised Modeling the radiative regime of glacier surfaces
an efficient and accurate BRF model that computes requires a quantitative understanding of how
surface reflectance without the need to evaluate the photons interact with complex surface composi-
light field within the particulate surface. Accurate tions. RT theory provides the physical basis for
numerical models that solve the RTE to determine modeling spectral and directional reflectance as a
both the light field within and reflected by the function of composition and granular texture. The
surface under investigation are also available. The RT problem and its mathematical machinery were
most popular model is DISORT (Stamnes et al. initially developed by Chandrasekhar (1960), who
1988), which is a generic RT code capable of simu- applied the model to the transfer of light in atmo-
lating the transport of photons in layered media spheres, oceans, and the interstellar medium. Here,
with specified composition and optical thickness. we are interested in detailing a modeling approach
A new multi-layer RT model based on recent theo- that yields RT equations capable of quantifying
retical and numerical advancements in solving the the light reflected and transmitted by glaciers and
RTE (Siewert 2000) has been developed and tested glacier lakes.
against currently available codes (Previti 2010; We begin by considering the general problem.
Previti et al. 2011; Picca et al. 2007, 2008a, Picca Sunlight is comprised of photons traveling through
2009). Whereas the new model is general enough to the atmosphere. Ultimately, they reach the Earth’s
model atmospheric RT, the Multi-layer Analytic surface and interact with media such as that of the
Discrete Ordinate Code (MADOC; Furfaro et al. surface of a glacier. The host medium is usually
2014) has been originally conceived to specifically characterized by an ensemble of particles with prop-
model the radiation regime within glacier environ- erties that are a function of their composition and
ments. spatial arrangement. Commonly, such ensembles of
This chapter focuses on a fundamental treatment particles include ice ranging from multicentimeter
of RT principles and the modeling of the radiation glacier ice to fine-grained snow and frost, liquid
reflected by glacier surfaces over the spectral range water, suspended particles in the water, and rock
between 0.4 and 2.5 mm (VIS/NIR). We emphasize debris or soot resting on the glacier or intermixed in
the processes required to compute surface reflec- the optical zone of the ice and snow. The ensembles
tance (e.g., BRF and BRDF) and spectral albedo, are modeled here as dense particulate media that
which are the most meaningful parameters in opti- are illuminated by both direct and diffuse beams of
cal remote sensing of Earth and planetary surfaces. photons and whose optical properties can be indi-
In Section 3.2, the principles of RT are introduced vidually described by a finite number of parameters.
and explained with special emphasis on a modeling Such parameters are intended to describe how the
approach to adapt the generic RTE to properly medium interacts with light particles. Generally,
describe the radiative regime for glacier surfaces photons are either absorbed or scattered. A single
and glacier lakes. In Section 3.3, we provide a photon can undergo multiple scattering before it is
thorough description of the optical properties of either absorbed by, or exits from the medium back
materials comprising a glacier surface (snow, ice, into the atmosphere and space. Modeling radiative
56 Radiative transfer modeling in the cryosphere

transport in dense particulate media is commonly Ss ðI 0 EIÞ ¼ ! ðÞp ð; cos YÞ where ! ðÞ and
done either stochastically or deterministically. p ð; cos YÞ are the single-scattering albedo and
Here, we will utilize a deterministic modeling the scattering phase function, respectively, and
approach that consists in formulating an integro- Y ¼ cos 1 ðI 0 EIÞ is the angle between the two
differential equation derived from a fundamental directions (see Section 3.2.2 for a more precise
law of physics (i.e., the balance of photons in the definition of both ! ðÞ and p ð; cos YÞ).
appropriate phase space). In classical RT theory, To complete the mathematical description of the
the medium is assumed to be a collection of dimen- balance of photons interacting with the host med-
sionless scattering and absorbing centers uniformly ium, proper boundary conditions that account for
distributed in a differential volume. The photons’ the radiative flux of photons entering the medium
behavior is determined by the probability of scatter- must be provided. Importantly, for the case of ice,
ing and/or absorption within the host medium, snow, debris, and their mixtures, the scattering pro-
assuming that streaming is possible between inter- cess does not alter photons’ energy (i.e., interaction
actions. If the conservation of photons is applied in with the host medium does not shift wavelength).
six-dimensional phase space (i.e., position and Therefore, photons maintain their energy while
velocity), the following equation is utilized: changing direction. Eq. (3.1) can then be solved
independently at each wavelength to determine
1 @I ðr; I; tÞ the radiative regime of the observed surface. In
þ IErI ðr; I; tÞ þ Stot ðrÞI ðr; I; tÞ
c @t supraglacial lakes, Raman scattering requires that
Z
1 photons change energy during their interaction pro-
¼ dI 0 Ss ðr; I 0 ! IÞI ðr; I 0 ; tÞ ð3:1Þ
4 cess. Consequently, eq. (3.1) must be modified with
an additional wavelength-dependent source term
where I ðr; I; tÞ is the spectral radiance (energy/m 2 that accounts for such effects on overall photon
sr s) of photons at location r traveling in the balance (see Section 3.2.3). Bioluminescence is nor-
direction I ¼ ð; ’Þ within the cone dI. Spectral mally absent or negligible and is ignored for glacial
radiance is the (unknown) physical quantity that lakes.
describes the light distribution within, entering, Eq. (3.1) is an integro-differential whose solution
and exiting the host medium. Importantly,  and has challenged researchers for decades. In the radia-
’ are cosines of the inclination angle and the azi- tive transfer and particle transport communities, it
muth angle, respectively, which are used to describe is known as the linearized Boltzmann equation
the photons’ direction of motion. The two terms on (LBE) and generally does not have any analytical,
the left-hand side of eq. (3.1) represent net energy closed-form solution. Although eq. (3.1) is the basis
loss of photons streaming out of the phase space for our modeling, several assumptions must be
which is balanced by energy loss due to the scatter- enforced to derive a more manageable form. One
ing and absorption (third term) and the inscattering immediate simplification comes from the observa-
of photons in the phase space (right-hand side). The tion that the 1=c term factoring the time derivative
participating medium is described by the absorption of I ðr; I; tÞ is smaller than the intensity flux time
and scattering coefficients. Here, Stot ðrÞ is the total rate (Davis and Knyazikhin 2005). The latter
interaction coefficient, defined as the sum of the implies that a steady state is reached almost instan-
absorption and scattering coefficient (Stot ðrÞ ¼ taneously. Indeed, radiative transfer in passive
Sabs ðrÞ þ Ssca ðrÞ). In addition, Ss ðr; I 0 ! IÞ is remote-sensing applications is generally modeled
the differential scattering coefficient (also called as a stationary phenomenon; time-dependent RT
differential scattering cross section or inscattering problems are considered only when modeling the
coefficient) which describes the probability that response of the surface to active pulsed illumination
photons traveling in the I 0 direction are scattered by remote-sensing instruments (e.g., LiDAR).
in the dI about I direction. In conventional RT
theory, the host medium is assumed to be ‘‘rota-
tionally invariant’’ (i.e., the differential scattering
coefficient depends only on the angle between I
3.2.1 RT modeling approach for
and I 0 ). With the latter position, and assuming that
glacier surfaces
the transport of photons is a function of only one
spatial variable (depth, ), it is customary to write Glacier surfaces and supraglacial lakes generally
the differential scattering as Ss ðI 0 ! IÞ ¼ exhibit a complex spatial arrangement of different
Radiative transfer modeling of glacier surfaces 57

materials. Consequently, any RT model should be Figs. 3.1(C–J) show that more complex situa-
flexible enough to permit the incorporation of tions, not well modeled as areal mixtures, are also
compositional and spatial structure scenarios that common on glaciers. Most glaciers possess broad
are typically found in such environments (e.g., a areas of impure ice stained by small amounts of
two-layer arrangement that includes an upper surficial or intermixed rock grit. As detailed close-
layer of mixed snow and debris overlaying a up images show (Figs. 3.1(I, J); J being an extreme
bottom layer of pure bubbly ice). The natural color stretch of I), the rock grit is so fine that it is
variability of surface composition and lake sus- optically thin and to some extent is intermixed with
pended debris, however, makes it extremely difficult ice in the upper few millimeters; this slight debris-
to develop a comprehensive RT model that is sim- laden ice has a structure of massive, clean, coarse
ple (i.e., mathematically tractable), accurate, and crystalline ice overlain by a zone just a couple of
computationally inexpensive. Clearly, if a compre- millimeters thick where ice is effectively intermixed
hensive description of the optical properties appear- with fine silicate rock particles. We refer to a homo-
ing in the LBE is available, one may attempt to geneous or random intermixing of two (or more)
directly solve eq. (3.1). While this is in principle phases, with one phase having high photon trans-
possible, the dimensionality of the problem (three mittance relative to the other phase, like an intimate
positions and two angular variables) coupled with mixture. Thus, intimate mixtures permit photons to
the integro-differential nature of the LBE make interact with two or more materials of differing
it virtually impossible to find accurate and fast optical properties (e.g., ice with intermixed rock
solutions. Consequently, the RTE must be flour). As we demonstrate in the simulation section,
simplified. such intimately mixed surfaces exhibit unique spec-
A closer look at the actual material and particle- tral properties and unique bidirectional reflectance
size arrangements typically found on glaciers characteristics different from pure ice, pure rock, or
provides the insight that simplified geometries can areal mixtures of rock and ice.
support rigorous but simplified RTEs to describe The glacier surface also has centimeter-scale grit-
the BRDF and spectral albedo. A vertically layered free ‘‘windows’’ that expose the coarse crystalline
approach implies that radiant intensity depends ‘‘blue ice’’, and optically thick centimeter-size rock
only on one spatial (optical depth) and two angular particles. Photons are admitted through the grit-
variables. The layering assumption is justified by free windows into coarse-grained massive ice below
real case examples found in natural glacier environ- the gritty ice. The grittier areas admit and emit
ments. For example, Fig. 3.1 depicts monomineralic fewer photons per square millimeter, but they dom-
and monolithologic materials, intimate mixtures, inate the area fraction of this surface. Although
and areal mixtures of materials on the Root and Figs. 3.1(I, J) point out the real world complexity
Kennicott Glaciers in the Wrangell Mountains, of such surfaces from a radiation transfer perspec-
Alaska. Whether a patch of surface material is effec- tive, we can approximate the geometry as a deep
tively monomineralic or monolithologic depends on layer of clean coarse crystalline ice overlain by a
the scale and the situation. To explain areal mix- thin layer of an intimate mixture of ice with fine
tures, consider Figs. 3.1(A, B). At the ASTER pixel rock detritus. The more complex real world geom-
footprint scale (15 m/pixel for VNIR), the major etry can then be simulated by patches of intimately
glaciological features (especially medial moraines mixed dirty ice with patches of pure clean ice and
and intermoraine lanes of ice, snow, and firn) would patches of pure debris. Complex, realistic scenarios
be totally resolved and are each optically thick; can therefore be generated in a stepwise fashion.
these materials may be approximated as either pure The layering assumption reduces the dimension
rock debris or pure medium-grained ice (firn). A of the RTE yet allows meaningful modeling of a
MODIS pixel (250 m) would not resolve the indi- wide range of spatially complex material structures,
vidual medial moraines and ice patches; in this case, including multi-layers of intimate mixtures. In this
linear areal mixing models could adequately simu- framework, two major configurations for material
late reflectance. So long as photons do not penetrate arrangement are possible. In the first configuration,
through grains far enough to interact with different intimate mixtures of ice and sediment can be mod-
types of minerals, the reflectance signature of eled as a homogeneous single layer (optically thick),
such patch-wise areal mixtures is simply the area- where the optical properties of the layer are derived
averaged reflectance spectra of the mineral compo- by a weighted average of the optical properties of
nents making up the surface. the pure components (see Section 3.3.4). In the sec-
58 Radiative transfer modeling in the cryosphere

Figure 3.1. (A–J) Examples of monomineralic and monolithologic materials, intimate mixtures, and areal mixtures
of materials on the Root and Kennicott Glaciers in the Wrangell Mountains, Alaska. On the right side of the figure are
illustrated a set of typical scenarios that can be modeled using multi-layer 1D, two-angle radiative transfer equations
(eq. 3.3). The geometrical configurations include single-layer and multi-layer arrangements of mixtures of ice and
rock debris. Figure can also be viewed as Online Supplement 3.1.

ond configuration, the surface is assumed to exhibit


3.2.2 Radiative transfer equation in
an idealized multi-layer structure. Each layer is
layered mixtures of snow, ice, and
modeled as homogeneous, with a defined optical
debris
thickness, and its composition may include either
a pure material or an intimate mixture of different The major assumption that reduces the problem
elements (ice, snow, debris). In addition, RT models dimension is to set the medium to be infinite in
should be able to describe scenarios where, in a the horizontal direction so that photon transport
single-layer configuration, the surface exhibits within the medium is one dimensional (the depth
patches of homogeneous material. Within the limits variable). If a single-layer configuration is consid-
of its spatial resolution, any remote-sensing instru- ered, a further simplification is obtained by working
ment measures the spectrally average electromag- with a spatially homogeneous medium. Vertical
netic radiation reflected by the area corresponding heterogeneity can be described by assuming a
to one pixel. Assuming an independent pixel multi-layer configuration where the medium is sub-
approximation (IPA), RT models can be devised divided in many layers, each having different optical
to compute the flux intensity of heterogeneous properties. Under these conditions, the LBE can be
patched areas by calculating the average flux inten- simplified. The spectral radiant intensity for a ver-
sity of homogeneous subpixel components weighted tically heterogeneous medium is governed by the
with their area coverage. following 1D, two-angle RTE (Chandrasekhar
Radiative transfer modeling of glacier surfaces 59

1960, Siewert 1978, 2000): radiation field (second term at the RHS). Eq. (3.5)
@ ensures that the continuity of radiant intensity at
 I ð; ; ’Þ þ I ð; ; ’Þ the interface between layers is respected. Eq. (3.6)
@ 
Z Z imposes a non-reentrant boundary condition at the
! ðÞ 2 0 1 bottom layer. Typically, the bottom layer is
¼  d’ d 0 p ð; cos YÞI ð;  0 ; ’ 0 Þ
4 0 1 assumed to be optically thick (i.e., semi-infinite).
ð3:2Þ
where, for a given wavelength ,R I ð; ; ’Þ is the 3.2.3 Radiative transfer equation in
spectral radiant intensity,  ¼ x0 tot dx 0 is the glacier lake waters
optical thickness (varying in the range  2 ½0; D),
RT modeling for glacier-related surface water can
 2 ½1; 1 is the cosine of the polar angle (meas-
be based on using a plane-parallel approximation,
ured from the positive -axis), ’ 2 ½0; 2 is the
where the water column is assumed to be layered,
azimuth angle, ! ðÞ ¼ sca ð; Þ=abs ð; Þ is the
and surface irradiance is assumed to be uniform
spatially dependent single-scattering albedo,
over the surface. Because of small horizontal varia-
defined as the ratio between scattering coefficient
tion in water optical properties, the RT problem
sca ð; Þ and total interaction (extinction) coeffi-
can be treated spatially using the depth variable.
cient (tot ð; Þ ¼ sca ð; Þ þ abs ð; Þ (i.e., sum
Using notation that is customary in the ocean optics
of absorption and scattering coefficient), and
community, radiance is governed by the following
p ð; cos YÞ is the spatially dependent scattering
integro-differential equation:
phase function.
In the case of a multi-layer approximation, the @
 Lðz; ; ’; Þ þ cðz; ÞLðz; ; ’; Þ
medium is subdivided into N layers each having @z
different optical properties. For each layer, the Z 2 Z 1
0
single-scattering albedo and the scattering phase ¼ bðz; Þ d’ d 0 ðz;  0 ! ; ’ 0 ! ’; Þ
0 1
function are defined either for a single material or
for a mixture, respectively as ! s , s ¼ 1; . . . ; N and  Lðz;  0 ; ’ 0 ; Þ þ Sðz; Þ; ð3:7Þ
p s ðcos YÞ, s ¼ 1; . . . ; N. For each of the N layers, where z is downward distance from the lake surface,
we write the following equation (s ¼ 1; . . . ; N):  is the cosine of the polar angle, and ’ is the
@ ðsÞ ðsÞ azimuth angle in the horizontal plane. The sum of
 I ð; ; ’Þ þ I  ð; ; ’Þ the scattering and absorption coefficient is called
@ 
ðsÞ Z 2 Z 2 the beam attenuation (extinction) coefficient
!
¼ d’ 0 ðsÞ
d 0 p  ðcos YÞI s ð;  0 ; ’ 0 Þ cðz; Þ ¼ aðz; Þ þ bðz; Þ, whereas ðz;  0 ! ;
4 0 0 ’ 0 ! ’; Þ is the scattering phase function. Proper-
ð3:3Þ ties aðz; Þ, bðz; Þ, and cðz; Þ are commonly
known as inherent optical properties (IOPs). IOPs
The RTE must be equipped with boundary con-
are critical parameters that describe the bulk optical
ditions describing radiant intensity at the top and
behavior of lake water and are of general interest in
bottom layers. Moreover, the continuity of radiance
inverse remote sensing. Sðz; Þ is a source function
between layers must be enforced. For s ¼ 1 (top
that accounts for the Raman scattering phenom-
boundary condition):
enon. Such a term complicates the nature of the
ð1Þ equation, and finding accurate numerical solutions
I  ð0; ; ’Þ ¼ S0 ð  0 Þð’  ’0 Þ þ f ð; ’Þ ð3:4Þ
becomes more challenging. This is due to the fact
At the interface between slabs: that in such cases the problem cannot be solved
9
ðsÞ ðs1Þ ‘‘one wavelength at a time’’ as in the conventional
I  ð ðsÞ ; ; ’Þ ¼ I  ð ðsÞ ; ; ’Þ =
ð3:5Þ case (see Section 3.2.2). The latter implies that in eq.
ð ðsþ1Þ ; ; ’Þ ;
ðsÞ ðsþ1Þ
I  ð ðsþ1Þ ; ; ’Þ ¼ I  (3.7), the wavelength variable is coupled with both
spatial and angular variables. Nevertheless, a sim-
For s ¼ N (bottom-boundary condition): plified version of eq. (3.7) which does not depend on
ðNÞ Sðz; Þ may be sufficient to describe photons’
I  ð0; ; ’Þ ¼ 0 ð3:6Þ
behavior because bioluminescence sources are very
where  2 ½0; 1. Eq. (3.4) represents the incident small during the daytime, and both fluorescence
directional field (first term on the RHS) and diffuse and Raman scattering become negligible.
60 Radiative transfer modeling in the cryosphere

Glacier lake water is generally assumed to be customarily described by a series of coefficients


either homogenous (well mixed) or vertically representing the projection of p ðcos YÞ on
stratified which, as in the case for glacier surfaces, Legendre polynomials. Generally, snow, ice, soil,
enables the use of multi-layer configuration models. and water exhibit a strongly forward-peaked scatter-
Eq. (3.7) is also equipped with boundary conditions ing function (i.e., relatively large single particles tend
that are similar to the one described in the previous to scatter radiation primarily in the forward direc-
section. The top layer is illuminated by an incident tion). The latter implies that a large number of
and diffuse source (see eq. 3.4). If lake depth is coefficients (of the order of thousands) are required
large, the medium is optically thick and non- to accurately characterize the single-particle phase
reentrant conditions at the bottom layer (eq. 3.6) function. The asymmetry parameter g, which is
are appropriate. If the lake is shallow, reflective defined as the cosine of the average of p ðcos YÞ
boundary conditions, which depend on lake bottom (also as the degree of forward scattering of the
composition, shall be imposed at the bottom layer, medium), may be conveniently employed to reduce
although we do not provide simulations of such the number of parameters required to describe the
instances. optical properties of a single particle. The popular
Heyney–Greenstein model is therefore widely used
to approximate the phase function for the particles
of interest:

3.3 OPTICAL PROPERTIES OF SNOW, 1  g2


pðcos YÞ ¼ ; g 2 ½1; 1
ICE, DEBRIS, MIXTURES, AND ð1  2g cos Y þ g 2 Þ 3=2
GLACIER LAKE WATER ð3:8Þ
The RTE can be solved as a function of surface/ The scattering phase function can be expanded in
water properties (e.g., mineralogical composition, Legendre polynomials Pl ðcos YÞ with coefficients
grain size distribution) and solar/sensor geometry. that are directly related to the asymmetry parameter
The RTE describes the multiple scattering of in the following manner:
photons through the host medium and critically
depends on single-particle optical behavior. From X
L

the photon–medium interaction point of view, such pðcos YÞ ¼ l Pl ðcos YÞ ð3:9Þ


l¼1
behavior is regulated by the ability of the host
medium to absorb and scatter light in any direction. l ¼ ð2j þ 1Þg l ð3:10Þ
Indeed, a single particle is characterized by absorp-
tion and scattering efficiencies that are related to Therefore, single-particle optical properties can be,
the probability that a photon will be scattered or at minimum, characterized by knowledge of single-
absorbed by a single particle of defined shape and scattering albedo and asymmetry parameter. Alter-
size. Scattering and absorption processes are natively, a number of scattering coefficients are
accounted for via two optical property parameters: required to describe the particle phase function.
(1) single-scattering albedo, which represents the Such parameters depend on the size of the particle,
probability of a single photon to be scattered in wavelength, index of refraction, and the shape of
any direction; and (2) the scattering phase function the particle. The most common approach to deter-
p ðcos YÞ (or ðz;  0 ! ; ’ 0 ! ’; Þ for glacier mine optical properties is to assume that the particle
lake water), with Y being the angle between the is a perfect sphere. In this case, optical properties
incident and scattering direction. The phase func- can be computed in closed form by directly solving
tion specifically describes the probability that a Maxwell equations. Mie theory (Wiscombe 1980)
photon coming from a direction I 0 ¼ ð 0 ; ’ 0 Þ is has been widely used to model the optical properties
scattered by a particle in direction I ¼ ð; ’Þ. Par- of snow, ice, and soil. More involved approaches,
ticulates that typically comprise surfaces in glacier such as the T-matrix method (Watermann 1971,
environments are rotationally invariant (i.e., the Mackowski and Mishchenko 1996) and the ray-
probability of scattering depends only on the rela- tracing method (Bohren and Barkstrom 1974,
tive angle between incident and scattered radiation). Macke et al. 1996), have been developed to account
While single-scattering albedo is a scalar parameter for arbitrary and irregular shapes. Such methods,
ranging between 0 and 1, the phase function is however, tend to be computationally expensive.
Optical properties of snow, ice, debris, mixtures, and glacier lake 61

3.3.1 Snow
From the radiative transfer perspective, snow can
be viewed as a collection of ice particles immersed in
air. A single particle of ice has variable shape and
size and an exact description of single-scattering
albedo and phase function involves using methods
of geometric optics. Yang and Liou (1998) applied
Monte Carlo–driven ray-tracing algorithms to com-
pute the optical properties of ice crystals for a sub-
stantial variety of shapes including plates, hollow
columns, bullet rosettes, and ice aggregates.
Whereas a database of properties has been gener-
ated, accurate description of the phase function for
such complex shapes implies that one needs RT
code capable of handling thousands of phase func-
tion coefficients to compute the radiative regime in
media with irregular particles, which results in
impractical computationally expensive calculations.
A more popular approach employed to describe
snow particle optical properties makes extensive
use of Mie-based code. Founded on Mie theory
(Wiscombe 1980, 1996), such code can effectively
compute efficiencies, single-scattering albedo, and
asymmetry parameters as a function of grain size.
However, calculations based on Mie theory may be Figure 3.2. Spectral behavior of the real part (A) and
severely limited by the fact that the theory works imaginary part (B) of the complex index of refraction for
rigorously only if the particles are assumed to be pure ice. Figure can also be viewed as Online Supple-
perfect spheres. Whereas single snow particles are ment 3.2.
not spheres, an ensemble of snow particles is pos-
tulated to behave as ‘‘optically equivalent spheres’’
increases with wavelength). Wiscombe and Warren
(i.e., a collection of snow grains having the same
(1980) studied the behavior of pure ice single-
volume/surface ratio; Dozier et al. 1987). Mugnai
particle optical parameters using Mie theory and
and Wiscombe (1980) demonstrated that a collec-
showed that the extinction coefficient and asym-
tion of nonoriented spheroids have the same scat-
metry parameter are relatively insensitive to wave-
tering behavior as an ensemble of spherical particles
length (typical value for g ranges between 0.88 and
of equivalent size. Mie-based code requires knowl-
1), and that ! (or the co-albedo 1  ! ) is mainly
edge of the complex index of refraction n ¼ nr þ ni .
responsible for the spectral variation of snow
Generally, the real part is interpreted as phase
albedo. Generally, ! is very close to 1 in the optical
velocity and the imaginary part describes absorp-
region (highly scattering snow medium across the
tion loss for an electromagnetic wave moving
visible) and decreases monotonically reaching a
through the particle. For the case of ice, the
minimum value of 0.5 in the NIR. Increasing grain
imaginary part is linked to volume absorption of
size is shown to decrease ! . Generally, values of 50
ice (Wiscombe and Warren 1980). Fig. 3.2 shows
mm are assumed for fresh snow, whereas 1 mm is
both real and imaginary parts of the index of refrac-
assumed for grain clusters or wet snow. Water in
tion for pure ice (Wiscombe 1980, Warren and
the snowpack is not usually modeled, as the index
Brandt 2008). Importantly, the magnitude of the
of refraction of water is very close to that of ice.
imaginary part varies across the visible and the
near-infrared (NIR) by many orders of magnitude
implying that: (1) ice is transparent in the visible
3.3.2 Glacier ice
region (small imaginary index of refraction); and
(2) ice is moderately absorptive in the NIR Whereas snow and ice are the major material com-
(imaginary index of refraction is larger and ponents, the physical state and the optical proper-
62 Radiative transfer modeling in the cryosphere

ties of various glacier surface characteristics under plied by the volume fraction of ice in the sample.
investigation vary dramatically. Snowfall is trans- Once the scattering and absorption coefficients are
formed to ice through a variety of mechanisms available, both the extinction coefficient and single-
including: (1) mechanical settling; (2) sintering; scattering albedo can be computed to complete the
(3) refreezing of meltwater; and (4) refreezing of optical characterization of the volumetric scattering
sublimating ice. The grain size of surface ice varies of clean ice.
widely. In the accumulation zone, fine snow may Microtopography is important on some glacier
dominate. In the firn zone (annealed snow), grain surfaces and can introduce complex scattering
size may effectively be in the range of several milli- and microshadowing effects. We do not address this
meters. In the ablation zone, bubbly ice may have important topic in this chapter.
an effective grain size of a few millimeters, but
dense, well-crystallized ice may have grain sizes of
3.3.3 Rock debris
1–10 cm. In some cases, glaciers may be effectively
layered in the optical zone, with a dusting of snow Debris-covered glaciers include varying amounts,
or bubbly ice overlying denser, coarser ice. Multiple grain sizes, and spatial arrangements of rock debris.
scattering of such complex structures requires Moraines may consist mainly of boulders and cover
proper modeling of the properties of single com- the ice completely. Debris patches may be scattered
ponents which are usually mixed with other com- amongst clean ice exposures. Fine rock flour may be
ponents (snow, bubbles, rock debris, and soot, see intimately mixed with ice. The optical characteriza-
Section 3.3.4) and possibly arranged in multi-layer tion of single-particle absorption and scattering for
configurations. However, glacier clean ice may be soil/sediment is, however, very difficult. Generally,
modeled as a collection of bubbles trapped within a soil particle distributions can vary in size, shape,
matrix of transparent ice. and mineralogy. Whereas Mie theory has been
The firn, which is snow material after the trans- fairly successful in describing the optical properties
formation process has begun, is initially porous and of snow and ice, it has not been able to accurately
contains interconnected air channels. As the density reproduce the optical properties of soil particles.
increases above 880 kg/m 3 , the channels close off Indeed most materials found on Earth and planet-
resulting in a mixture of ice and bubbles trapped ary surfaces exhibit a nonuniform structure and
within the glacier body. Mullen and Warren (1988) composition.
provided a framework for modeling the volumetric Two decades ago, Hapke (1993) presented a gen-
scattering of such bubbly ice. Indeed, in a pure ice eralized model for the scattering efficiency of large
sample containing only air bubbles, the physics of irregular particles. The model is based on an exten-
interaction between the photons and the host med- sion/generalization of the equivalent slab model for
ium is such that absorption occurs in the ice matrix irregular particles. The model is approximate but
and scattering occurs at ice bubble boundaries. can be used to compute scattering efficiency in
Thus, the absorption process is generally separated closed analytical form and has been mostly
from the scattering process. Scattering is dominated employed in the planetary science community
by the size and distribution of the air bubbles within (e.g., Warell et al. 2009). More recently, T-matrix
the ice. If bubbles are assumed to be spheres, Mie code has been made available to describe the optical
theory can be employed to compute the scattering properties of particles that are large and irregular.
efficiency (and subsequently the scattering coeffi- Based on the linearity of Maxwell’s equations
cient) as well as the asymmetry parameter as a (Watermann 1971), the method has some computa-
function of bubble size. In this case, the Mie-based tional advantage where the particle under consid-
calculation follows the same process as in the case eration is axial symmetric, because the matrix is
of snow, but with the imaginary part of the complex easily subdivided into independent matrices, yield-
of refraction set to zero (n ¼ nr ). Marston et al. ing more efficient and faster calculations. T-matrix
(1982) found good agreement with the scattering methods are also available to compute the optical
of a single air bubble in water, which has an index properties of particle clusters with defined
of refraction similar to ice. Conversely, absorption orientation (Mackowski and Mishchenko 1996).
is assumed to be exclusively a function of the Generally, such algorithms are computationally
amount of ice per unit volume. For example, expensive, and recently some of the available code
Bohren (1983) assumed that the absorption coeffi- has been redesigned to run on parallel clusters of
cient of an air–ice mixture is that of pure ice multi- machines (Mackowski and Mishchenko 2011).
Optical properties of snow, ice, debris, mixtures, and glacier lake 63

3.3.4 Mixtures environments ranging from exquisite turquoise and


other pastels, to brilliant green, cobalt blue, and
The optical properties of single particles can be
gray. The colors of optically deep lakes pertain
employed to determine the optical behavior of
mainly to their suspended sediment loads, especially
multicomponent mixtures. Computing single-
clastic sediment in glacial lakes and substrate effects
scattering albedo and the asymmetry parameter
for optically shallow lakes.
for a mixture is fairly straightforward. Let us
Suspended clastics in glacier meltwater have two
consider the case of two pure components each
major components: (1) rock flour—mainly clay to
characterized by spherical particles of radius r1
fine silt size—produced by glacier grinding at the
and r2 and with scattering coefficient sca;1ð2Þ ¼
bed; (2) silt and sand introduced by nonglacial
n1ð2Þ r 21ð2Þ Qsca;1ð2Þ and total interaction (extinction)
streams and incorporated into glacier waters from
coefficient tot;1ð2Þ ¼ n1ð2Þ r 21ð2Þ Qtot;1ð2Þ where
mass-wasting deposits or other sources that have
Qsca;12 and Qtot;12 are the scattering and extinc-
not been pulverized at the glacier bed. These com-
tion efficiencies for type 1(2) particles, and n1ð2Þ is
ponents are usually distinct in their grain-rounding,
the number of particles per unit volume of type 1(2)
faceting, and particle size–frequency distributions.
particles. The combined single-scattering albedo of
In heavily glacierized basins (where glacial sediment
a mixture of two types of particles can be deter-
input is much greater than nonglacial input) in
mined as follows:
nonpolar climates (where glaciers have thawed
sca;1 þ sca;2 n1 r 21 Qsca;1 þ n2 r 22 Qsca;2 beds), glacier rock flour overwhelmingly dominates
!mix ¼ ¼ most lakes. Glacial grinding produces extremely
tot;1 þ tot;2 n1 r 21 Qtot;1 þ n2 r 22 Qtot;2
fine particles due to rock being forced to slide over
ð3:11Þ other rocks, whereas in ordinary streams, particles
If particles are not spherical, r 2 shall be replaced are generated as chips that are shed from larger
by relative particle cross-sectional area. If particle clasts during water-driven impacts. The energy
size is modeled as a size distribution, then comput- and momentum relations are such that particles a
ing an integral with the weighted size distribution few microns in radius become ineffective abrading
is required. Similarly, combined phase function agents in normal streams. Hence, glacial rock flour
moments are computed by averaging the moment includes particles sizes from hundreds of microns to
of each component and weighting it by its scattering a tenth of a micron or smaller, whereas non-glacial
coefficient. For example, the asymmetry parameter streams normally carry suspended load particle
of a two-component mixture is computed as fol- sizes from a few hundred microns down to a few
lows: microns. Although particle sizes on the scale of light
wavelengths can total up to a small fraction of the
g1 sca;1 þ g2 sca;2
gmix ¼ total mass of suspended load, those size ranges can
sca;1 þ sca;2 dominate scattering due to the large specific surface
g1 n1 r 21 Qsca;1 þ g2 n2 r 22 Qsca;2 area of those fine grains. Such particle sizes can
¼ ð3:12Þ be virtually absent from nonglacial waters, thus
n1 r 21 Qsca;1 þ n2 r 22 Qsca;2
accounting for some of the difference between the
colors of typical nonglacial lakes versus glacial
lakes.
3.3.5 Glacier lake water
From the optical perspective, turbid nonglacial
Glacier water includes lakes, streams, and small lake waters, which are generally classified as ‘‘Case
puddles and rivulets. In this section we will not 2’’ waters (Mobley 1994), can be difficult to
consider thin films (such as wetted rocks and characterize because they often consist of water
water-saturated snow) or optically thin ponds plus dissolved organic and inorganic matter. Most
where substrate effects are important. The optical glacial lakes and streams have negligible organic
properties of optically thick glacier water are components, but some isolated ponds can develop
described by explicitly characterizing the absorp- a biota and suspended and dissolved organic mat-
tion, scattering, and total beam attenuation coeffi- ter. These lakes are strongly influenced by colored
cients as a function of the various components dissolved organic matter (CDOM also known as
present in the water body. Optical properties are Gelbstoff, yellow substance, or gilvin), detritus,
responsible for the large variation in the supra- mineral particles, bubbles, and other substances
glacial lake colors that can be observed in alpine including phytoplankton. Such particles are not
64 Radiative transfer modeling in the cryosphere

and directly proportional to its concentration to the


power of 0.6. Suspended particle optical properties
in glacier lakes are not well known and require in
situ measurement. Suspended sediments tend to be
mostly scatterers although iron-rich sediments may
exhibit strong absorption features. Heege (2000)
measured the optical behavior of suspended par-
ticles in Lake Constance and found that the par-
ticles had no influence on the overall absorption
coefficient (aX ðÞ ¼ 0) whereas they heavily influ-
enced the scattering properties of the lake waters. It
was found that the backscattering coefficient was
independent of wavelength and proportional to
the particle concentration.
Scattering in glacier and oceanic environments is
Figure 3.3. Spectral behavior of absorption and scat- strongly forward peaked with peak ratio between
tering coefficients for pure water. Figure can also be forward and backward single-scattering ratios of
viewed as Online Supplement 3.3. the order of 10 5 . Various models are available to
describe such a type of scattering behavior although
two types of phase functions are generally used for
water including the Petzold phase function (Petzold
very well characterized in terms of type and loca-
1972) and the Fournier–Fourand model (Fournier
tion, and they do not covary with phytoplankton
and Fourand 1994). Mobley et al. (2002) examined
concentration as in Case 1 waters. Generally, the
the effect of various phase functions on the com-
optical properties of such waters are assumed to be
puted radiance and concluded that: (1) knowledge
additive, where the absorption and scattering coeffi-
of the appropriate phase function is as important
cient can be written as a sum of the models
as knowledge of the absorption and scattering
employed for absorption and scattering of the indi-
coefficient; and (2) accurate knowledge of the shape
vidual components:
of the phase function is not required as long as
the backscattering fraction is correct. It was also
aðÞ ¼ aW ðÞ þ aP ðÞ þ aX ðÞ þ aY ðÞ ð3:13Þ
shown that the Fournier–Fourand model works
bðÞ ¼ bW ðÞ þ bP ðÞ þ bX ðÞ þ bY ðÞ ð3:14Þ very well.

where aW ðÞ; aP ðÞ; aX ðÞ; aY ðÞ (bW ðÞ; bP ðÞ;


bX ðÞ; bY ðÞ) are the absorption (scattering) coeffi-
cients for pure water, phytoplankton, mineral par-
ticles/detritus, and CDOM, respectively. Models 3.4 NUMERICAL SOLUTION OF
and/or experimental data for individual compo- THE RTE
nents are more or less available. For pure water,
scattering and absorption coefficients have been Computing the radiative regime within glaciers, as
measured. Fig. 3.3 shows the wavelength-dependent well as the amount of radiation reflected by glacier
scattering and absorption coefficient of pure water surfaces and/or glacier lakes, requires solving the
as measured by Pope and Fry (1997). Optical RTEs formally presented in the previous sections.
models for phytoplankton and CDOM are avail- Due to its mathematical complexity, an analytical
able from the literature (Prieur and Sathyendranath description of the light field is virtually impossible.
1981). Generally, CDOM absorption is assumed to Over the past few years, many approximate meth-
decay exponentially with wavelength, whereas scat- ods have been developed to provide analytical
tering due to dissolved microscopic pigments is expressions for the multiple scattering of photons
assumed to be negligible. Phytoplankton absorp- in snow, ice, and soil. One of the most popular
tion is a complex function of its concentration analytical methods is based on approximating the
and wavelength, and requires two empirical wave- radiance as one upwelling and one downwelling
length-dependent functions. Phytoplankton scatter- stream of photons as they are constrained to move
ing is proportional to the inverse of the wavelength only in two directions (two-stream approximation).
Numerical solution of the RTE 65

Such an assumption allows the determination of have unknown radiant intensities along discrete
analytical formulas that quickly compute upwelling directions. The set of equations are solved by:
and downwelling diffuse fields (see Wiscombe and (1) numerically solving the resulting eigenvalue
Warren 1980 for an example of a two-stream calcu- problem to compute the homogeneous solution;
lation applied to the problem of determining snow and (2) use of a modified Green’s function formula-
albedo). In an attempt to provide a quantitative tion to compute the particular solution of the set of
tool for the analysis and interpretation of the light differential equations arising from angular dis-
reflected by planetary surfaces, Hapke (1981) deter- cretization. More recently, the method has been
mined an analytical expression for the reflectance extended to include multiple layers of optical prop-
factor of an optically thick surface. He decomposed erties with special routines that give the method the
the radiative field into two major components con- ability to quickly and efficiently handle thousands
sisting of a scattered component and a multiple- of layers (Picca et al. 2007, 2008a, Picca 2009, Pre-
scattered component. An approximate solution viti 2010, Previti et al. 2011). The principles behind
was obtained by: (1) determining an exact expres- the ADO have been implemented in a novel RT
sion for the first scattered component; and (2) deter- code called Multi-layer Analytical Discrete Ordi-
mining an approximate analytical expression for nates Code (MADOC, Furfaro et al. 2014). Devel-
the multiple-scattered component, assuming that oped in a MATLAB environment, MADOC will be
the host medium is isotropic from the second col- soon made available to the larger scientific com-
lision on. munity. Importantly, MADOC was used to provide
Whereas approximate methods provide the the simulation examples in the next section. Dis-
analyst with analytical expressions that are easy crete ordinate methods have also been used to gen-
to implement, they do not provide accurate solu- erate code capable of computing matter/energy
tions of the RTE. Indeed, accurate solutions can be interactions in water. Indeed, the Coupled Ocean
found only by solving the RTE numerically. Over Atmosphere Radiative Transfer code (COART; Jin
the past two decades, the widespread availability of et al. 2006) extended the DISORT numerical plat-
high-speed digital computers permitted researchers form to handle the change in photon direction
to develop and test more efficient and faster algo- between the water and atmosphere interface due
rithms to compute radiance and the reflectance fac- to differences in the index of refraction.
tor. All the available numerical algorithms have In principle, both angular and spatial variables
been based on some form of discretization of the can be discretized to transform the RTE into a set
spatial and angular variable, thus resulting in RT of differential equations that can be implemented
software that can be used to simulate the light using a digital computer. The most elementary
reflected by various surfaces. approach to what is called ‘‘vanilla discretization’’
One of the most efficient ways to solve the RTE is is the SN1 method (Lewis and Miller 1984). One
to discretize the angular variable in a set of finite major drawback of the method is that accurate
directions, and then solve the resulting differential solutions can only be obtained if the discretization
equations. Cast as an eigenvalue problem, the set of mesh is sufficiently fine. For optically thick media
equations can be solved using conventional numer- and settings with forward-peaked phase functions,
ical methods (e.g., relaxation method) to find an the method becomes computationally expensive.
analytical expression for radiance along fixed direc- More recently, it has been shown that accurate
tions. The DISORT code (Stamnes et al. 1988) is solutions of the RT equation can be obtained by
the most popular and is based on discretizing the solving a sequence of RT problems using SN from
angular variable. It has been widely used by the coarser to finer grids and coupled with acceleration
community to compute the reflectance factor and techniques to find the mesh-independent limit
spectral albedo of snow as a function of grain size (Ganapol and Furfaro 2008).
(e.g., Aoki et al. 2000, Painter et al. 2003). Two other alternative methods are suitable for
Recently, Siewert (2000) proposed a new formu- RT calculations involving particulate media. The
lation of the discrete ordinate approach for a single ‘‘adding–doubling method’’ (Hansen and Travis
layer with specified optical properties and thickness. 1974) is based on the idea that the medium is sliced
The method, called the analytical discrete ordinates in thin layers with light entering the top layer. The
method (ADO), is a novel semi-analytic approach angular reflectance (and transmittance) of the com-
in which the angular variable is discretized to deter-
1
mine a set of ordinary differential equations that SN is the codeword for discrete ordinates method.
66 Radiative transfer modeling in the cryosphere

Figure 3.4. RT-based model of spectral albedo for a layer of optically thick pure snow as a function of wavelength
(0.4–2.5 7m) and grain size (50, 200, and 1,000 7m). The solar zenith angle is 30 . Compare with Figure 2.8 (Bishop
et al.’s chapter). Figure can also be viewed as Online Supplement 3.4.

bined layers is obtained by superposing the reflec- 3.5 GLACIER RADIATIVE TRANSFER
tance and transmittance of the individual layers. SIMULATION EXAMPLES
The adding–doubling model has been implemented
by de Hann et al. (1987). Another efficient RT Proper modeling of the BRF and spectral albedo
model devised to compute the bidirectional reflec- based upon surface properties requires solving the
tance factor of an optically thick surface of parti- RTE as presented in eqs. (3.3) and (3.7). Here, we
culate media (including snow, ice, debris, and recall their definitions (see Chapter 2 of this book
mixtures) was implemented by Mishchenko et al. by Bishop et al.). The bidirectional reflectance fac-
(1999). The method is based on iteratively solving tor (BRF) represents the ratio between the radiance
a nonlinear integral equation (an Ambartsumian emitted in any particular direction by the surface,
equation) derived from photon conservation. The and the radiance that would be reflected into the
method is efficient because it does not need to solve same direction given an ideal Lambertian surface,
for light inside the medium and may be ideal for illuminated by the same incident geometry. Con-
modeling reflected radiance. versely, the spectral albedo of a surface is a dimen-
Finally, the light field in water can be computed sionless ratio of the radiant energy scattered away
using a commercially available program called from the surface to that received by the surface at a
HYDROLIGHT (Mobley 1994). Written in For- specified wavelength. In this section, a set of numer-
tran, the code uses the invariant-embedding ical examples that show how RT theory can be
approach to solve eq. (3.7) for each of the azimuthal employed to model BRF and spectral albedo for
components. It has an extensive library of optical a variety of configurations typically found in alpine
properties, and may be purchased for a fee. glaciers and glacier lakes are presented. The
Glacier radiative transfer simulation examples 67

Figure 3.5. RT-based model of spectral albedo for a layer of optically thick pure snow as a function of wavelength
and solar zenith angle. The particle grain size is fixed and assumed to be 50 7m. Figure can also be viewed as Online
Supplement 3.5.

MADOC platform (Previti 2010, Previti et al. 2011, photons have higher chances to be scattered at
Furfaro et al. 2014) has been employed in all simu- the boundary between fine snow grains and air.
lations involving glacier surfaces. Conversely, para- Incrementally larger grain size has the effect of
meterization of HYDROLIGHT, as presented by increasing the mean free path, giving photons a
Albert and Mobley (2003), has been employed to higher chance to travel through the ice, and a
model reflectance factors for glacier lake scenarios. smaller chance to be scattered and exit the snow-
In the first set of simulations, RT theory has been pack. Whereas larger snow particles are both more
employed to model the spectral albedo of a layer of absorptive and more forward scattering, it can be
optically thick pure snow as a function of wave- shown that the decrease in albedo is mainly due to
length (0.4–2.5 mm) and grain size (Fig. 3.4). The the fall of ! in the NIR regime, where the asym-
solar zenith angle was set at 30 . The single-particle metry parameter increases only slightly (Wiscombe
optical properties of single-scattering albedo and and Warren 1980). Importantly, in the visible
asymmetry parameter, were computed using the region of the spectrum, snow particles are highly
MATLAB-based Mie code provided by Matzler scattering (! very close to one), which explains
(2004). The ice complex index of refraction was why snow has generally such a high albedo, inde-
used as input and grain sizes of 50 mm (fresh fine pendent of grain size.
snow), 200 mm (fresh coarse snow) and 1 mm In the second set of simulations, a RT model was
(annealed snow; i.e., firn) were considered. employed to compute the spectral albedo of a layer
As shown in Fig. 3.4, albedo is very sensitive to of optically thick pure snow as a function of wave-
grain size and decreases as the radius of the snow length (0.4–2.5 mm) and solar zenith angle (Fig. 3.5).
particle increases. From the physical point view, Grain size was fixed and assumed to be 50 mm. Mie
68 Radiative transfer modeling in the cryosphere

Figure 3.6. RT-based model of spectral albedo for a layer of an optically thick mixture of pure snow and carbon
soot as a function of wavelength and snow grain size (50, 200, and 1,000 7m). Soot particle concentration is
assumed to be 3 ppmw. The optical properties of the two constituents have been independently computed using
Mie theory. Figure can also be viewed as Online Supplement 3.6.

theory is used to compute single-particle optical radius equal to 0.1 mm and particle density equiva-
properties. As expected, the angle of incidence of lent to 0.3 ppmw (parts per million weight). The
the incoming radiation illuminating the snowpack soot complex index of refraction is assumed to be
has a large effect on albedo. As shown in Fig. 3.5, constant across the spectral range of interest and
albedo increases with increasing solar zenith angle. equal to 1:95 þ 0:79i. Fig. 3.6 shows the resulting
Wiscombe and Warren (1980) explained this phe- spectral albedo as a function of wavelength along
nomenon by postulating that, because of the high with the radius of the snow particles. Simulations
inclination with respect to the zenith, photons show that a small amount of carbon soot is suffi-
entering the medium travel close to the upper sur- cient to lower albedo in the visible region of the
face of the snowpack and therefore scattering spectrum where ice has the lowest absorption
events give light particles a higher probability of (highly scattering media with single-scattering
exiting the snowpack surface. albedo very close to one). As expected, albedo
In the third set of simulations, a RT model was reduction is more marked for larger snow particles.
employed to compute the spectral albedo of a layer Relatively high scattering between 0.4 mm and 0.7
of an optically thick mixture of pure snow and mm increases the probability that photons are more
carbon soot as a function of grain size (50, 200, likely to experience multiple scattering, therefore
and 1,000 mm; see Fig. 3.6). The optical properties increasing the probability of encountering a carbon
of the two constituents have been independently particle and being absorbed. As discussed above,
computed using Mie theory. The optical properties increasing snow particle grain size increases average
of carbon soot were computed (assuming the soot free mean path, further increasing the probability of
particles are modeled as spheres) by setting the soot encountering a carbon particle and being absorbed.
Glacier radiative transfer simulation examples 69

Table 3.1. Input optical parameters employed for become more spatially homogeneous. The forward-
MADOC BRF simulations of intimate and areal scattering component also decreases in extent and
mixtures of ice and soil. magnitude, while the backscatter component
Parameters Ice Soil decreases in extent, and exhibits an increase in mag-
nitude associated with increasing debris.
Wavelength (mm) 0.63 0.63 In the fifth set of simulations, the reflectance
patterns of typical glacier lake water were computed
Particle diameter (mm) 30 10 as a function of particle and phytoplankton con-
Variance 0.2 0.1 centration, wind speed, and sensor geometry. Fig.
3.8 depicts spectral curves from 390 to 710 nm.
Distribution Log Modified Simulations were implemented using the analytical
normal gamma fit of HYDROLIGHT for Case 2 (turbid) waters
(Albert and Mobley 2003). Clearly, increasing
Index of refraction (real) 1.31 1.55
the suspended matter content tends to decrease
Index of refraction (imag) 0 0.001 reflectance variability (flatten spectral curves) and
increase the magnitude of reflectance from the lake
Max radius 35 N/A

Min radius 25 N/A

Single-scattering albedo 1 0.85413

Max scattering coefficients 641 641

Type of simulation Intimate Intimate


mixture mixture

In the NIR region (above 0.9 mm) the influence of


carbon particles on albedo is limited, as its reduc-
tion is dominated by the stronger absorption of ice.
This simulation shows that carbon soot and gener-
ally other impurities (e.g., dust; Wiscombe and
Warren 1980) may have a large impact on the over-
all energy budget of glaciers.
The fourth set of simulations shows an example
of a BRF for an optically thick mixture of soil and
ice as a function of the percentage volume of the
two pure components. The mixture BRF was com-
puted at a specified wavelength in the visible region
of the spectrum (0.63 mm). Input parameters,
including the complex index of refraction, grain
size, and grain size distribution for both soil and
snow are reported in Table 3.1 (Warren 1984, Mish-
chenko et al. 1999). A total of 641 coefficients of the
Legendre expansion were used to properly describe
the phase function. Fig. 3.7 demonstrates the reflec-
tance variability associated with intimate mixtures Figure 3.7. BRF simulations for intimate mixtures (in
volume percentage) of ice and sediment/soil. The over-
of ice and sediment/soil. Clearly, the magnitude of
all magnitude (i.e., albedo) decreases as the percentage
reflectance decreases as the percentage of sediment/
of soil increases. BRF patterns are also a function of the
soil increases. The anisotropic nature of reflectance mixture percentage. This set of simulations shows that
is also noteworthy, as pure ice exhibits a highly BRF patterns can be potentially used to discriminate
variable pattern that is directionally dependent. between various surface materials and conditions in
The anisotropic pattern changes with increasing glacier environments. Figure can also be viewed as
sediment, such that azimuthal reflectance variations Online Supplement 3.7.
70 Radiative transfer modeling in the cryosphere

Figure 3.8. Simulated spectral reflectance curves from 390 nm to 710 nm as a function of suspended particles and
phytoplankton concentrations, wind speed, and Sun angle. Simulations are based on an analytical fit of the
HYDROLIGHT radiative transfer model adapted to simulate Case 2 water. (A) Effect of suspended matter on
deep-lake water reflectance. As concentration increases, spectral variability decreases, although the magnitude of
reflectance increases across the wavelength range. (B) Effect of phytoplankton, causing increased reflectance with
a peak in the green region of the spectrum. (C) Effect of wind speed (u) on spectral reflectance. A decrease in wind
speed results in higher reflectance. (D) Effect of solar zenith angle on reflectance. The range of angles causes a small
change in reflectance but no significant change in absorption. Figure can also be viewed as Online Supplement 3.8.

water surface. In fact, as the particle concentration 3.6 CONCLUSIONS


increases, scattering dominates over absorption. In
addition, measurements over Lake Constance water Radiative transfer theory is a powerful tool that can
(Gege 1998, Heege 2000) showed that scattering be employed to quantify the amount of radiation
(and backscattering) is fairly insensitive to wave- reflected by glacier surfaces and glacier lake water.
length. In most cases, glacier lake turbidity is so The equations of radiative transfer, which describe
high that large optical depths effectively preclude the balance of photons that are being absorbed and
transmittance of light that has reflected off the lake scattered in their interaction with the host medium,
bottom, except in uncommon cases of extremely are mathematically complex. Nevertheless, over the
shallow lakes and areas fringing the edges of deep past few decades, advancements in numerical mod-
lakes. The more usual bathymetry of glacier lakes eling of RTEs have provided the scientific commun-
causes very narrow edge zones where the bottom is ity with RT software that permits the computation
sensed, so to a good approximation one can com- of physical quantities of interest in remote sensing
pletely ignore bottom reflectance and assume a (e.g., BRF and spectral albedo) both efficiently and
condition of infinite optical thickness. However, accurately. Such RTE-based radiance calculations,
by modifying bottom boundary conditions (i.e., however, are only meaningful if knowledge of the
assuming that the bottom of the lake is reflecting), medium’s optical properties is available. Such
RT code for water can be employed to compute the RTEs require the input of explicit models for
spectral reflectance of shallow lakes. Importantly, single-scattering albedo and scattering phase func-
such simulations require guestimating the spectral tion, which in turn depend upon the use of ab-
reflectance properties of the lake bottom. sorption and scattering coefficients (as well as
References 71

efficiencies), the asymmetry parameter (or Legendre depend continuously on the data. Recent mathe-
coefficients describing series expansion of the phase matical advancements in the field of statistics
function), particle concentrations, wavelength of applied to machine learning (e.g., neural networks,
interest, as well as size and shape of the single-type Gaussian processes, regularized sliced inverse
particles/mixture of many-type particles describing regression), however, have made model inversion
the medium. Mie theory is the most popular way to more manageable. Examples of such approaches
determine such properties. can be found in the literature for a variety of appli-
Based on the assumption that a single particle is cations in Earth and planetary science (e.g., Furfaro
spherical, Mie-based code has been designed to et al. 2007, Morris et al. 2008, Bernard-Michel et al.
compute the aforementioned parameters as a func- 2009). The latter reinforce the significance of RT
tion of the complex index of refraction and particle modeling for studying and understanding the
size. The spherical particle assumption has been Earth’s cryosphere.
shown not to be limiting for snow and ice mixtures,
but does not accurately model soil/sediment optical
properties. More advanced techniques based on the 3.7 REFERENCES
T-matrix and ray-tracing methods have been
designed to model the optical properties for par- Albert, A., and Mobley, C.D. (2003) An analytical model
ticles of irregular shape. Nevertheless, they tend for subsurface irradiance and remote sensing reflec-
to be computationally expensive. tance in deep and shallow case-2 waters. Optics
The optical properties of glacier lake water have Express, 11(22), 2873–2890.
been modeled by experimentally determining the Aoki, T., Aoki, T., Fukabori, M., Hachikubo, A.,
Tachibana, Y., and Nishio, F. (2000) Effects of snow
scattering and absorption coefficient of the various
physical parameters on spectral albedo and bidirec-
components comprising the host medium (pure tional reflectance of snow surface. Journal of Geophys-
water, phytoplankton, particle concentration, ical Research, 105, 10219–10236.
CDOM). Models for a scattering phase function Bernard-Michel, C., Douté, S., Fauvel, M., Gardes, L.,
that is suitable to describe the in-scattering of Case and Girard, S. (2009) Retrieval of Mars surface phys-
2 waters has also been proposed. ical properties from OMEGA hyperspectral images
The RT simulation examples have demonstrated using regularized sliced inverse regression. Journal of
that modeling can be effectively employed as an Geophysical Research (Planets), 114, 6005.
investigative tool to assess the information content Bishop, M.P., Barry, R.G., Bush, A.B.G., Copland, L.,
in satellite imagery. Furthermore they show how Dwyer, J.L., Fountain, A.G., Haeberli, W., Hall,
typical glacier surface conditions and glacial lakes D.K., Kääb, A., Kargel, J.S. et al. (2004) Global
optically respond as a function of their morpholo- land-ice measurements from space (GLIMS): Remote
sensing and GIS investigations of the Earth’s cryo-
gical and mineralogical composition. Consequently,
sphere. Geocarto International, 19(2), 57–84.
the coupling of RT modeling and multispectral digi-
Bohren, C.F. (1983) Colors of snow, frozen waterfall and
tal image analysis can be used to assess important icebergs. J. Opt. Soc. Am., 73, 1646–1652.
biophysical parameters of glacier environments, Bohren, C.F. and Barkstrom, B.R. (1974) Theory of the
such that the assimilation of sensor data can pro- optical properties of snow. Journal of Geophysical
duce spatiotemporal quantitative information Research, 79, 4527-4535.
regarding the observed medium including ice grain Bush, A.B.G. (2000) A positive feedback mechanism for
size, percentage of carbon soot in snow, concentra- Himalayan glaciation. Quaternary International, 65(6),
tion of glacier flour in lakes, pixel debris cover 3–13.
percentages. Such attempts or algorithms are Chandrasekhar, S. (1960) Radiative Transfer, Dover,
usually called ‘‘physically based inversion algo- New York.
rithms’’, and they are essentially models based on Davis, A.B., and Knyazikhin, Y. (2005) A primer in 3D
radiative transfer. 3D Radiative Transfer in Cloudy
first principles which are employed to describe the
Atmospheres, Vol. 2, Springer-Verlag, Berlin, pp.
functional inverse relationship between a sensor-
153–242.
collected signal and the properties of the observed de Hann, J.F., Bosma, P.B., and Hovenier, J.W. (1987)
surface/medium (Furfaro 2004, Furfaro et al. 2007, The adding method for multiple scattering calculations
Picca et al. 2008b). It is important to note, however, of polarized light. Astron. Astrophys., 183, 371–391.
that such RT and coupling issues can be difficult Dozier, J., Davis, R.E., and Perla, R. (1987) On the
problems to solve. Indeed, it is well known that objective analysis of snow microstructure. In: B. Salm
inverse problems are ill posed in that they do not and H. Gubler (Eds.), Avalanche Formation, Movement
72 Radiative transfer modeling in the cryosphere

and Effects (IAHS Pub. No. 62), International Associa- transfer in the coupled atmosphere–ocean system with
tion of Hydrological Sciences, Wallingford, U.K., pp. rough surface. Applied Optics, 45, 7443–7455.
49–59. Kargel, J.S., Abrams, M.J., Bishop, M.P., Bush, A.,
Fournier, G.R., and Fourand, J.L. (1994) Analytical Hamilton, G., Jiskoot, H., Kääb, A., Kieffer, H.H.,
phase function for ocean water. In: J.S. Jaffe (Ed.), Lee, E.M., Paul, F. et al. (2005) Multispectral imaging
Ocean Optics XII (Proc. SPIE 2258), Society of contributions to global land ice measurements from
Photo-Optical Instrumentation Engineers, Bellingham, space. Remote Sensing of the Environment, 99, 187–219.
WA, pp. 194–201. Kotlyakov, V.M., Serebrjanny, L.R., and Solomina,
Furfaro, R. (2004) Radiative transport in plant canopies: O.N. (1991) Climate change and glacier fluctuations
Forward and inverse problem for UAV applications. during the last 1,000 years in the southern mountains
Ph.D. Dissertation, University of Arizona. of the USSR. Mountain Research and Development, 11,
Furfaro, R., Ganapol, B.D., Johnson, L.F., and Herwitz, 1–12.
S.R. (2007) Coffee ripeness monitoring with airborne Lewis, E.E., and Miller, W.F. (1984) Computational
imagery: Neural network algorithm for field ripeness Methods of Neutron Transport, Wiley-Interscience,
evaluation using reduced a priori information. Applied New York.
Engineering in Agriculture, 23, 379–387. Macke, A., Mueller, J., and Raschke, E. (1996) Single
Furfaro, R., Previti, A., Picca, P., Mostacci, D., Ganapol, scattering properties of atmospheric ice crystals. Jour-
B.D. (2014) Multi-layer Analytic Discrete Ordinate nal of Atmospheric Sciences, 53(19), 2813–2825
Code (MADOC): An accurate and efficient radiative Mackowski, D.W., and Mishchenko, M.I. (1996) Calcu-
transfer platform to compute the bidirectional reflec- lation of the T matrix and the scattering matrix for
tance factor and spectral albedo of highly inhomo- ensembles of spheres. J. Opt. Soc. Am. A, 13, 2266–
geneous planetary surfaces. Journal of Quantitative 2278.
Spectroscopy and Radiative Transfer (in preparation). Mackowski, D.W., and Mishchenko, M.I. (2011) A mul-
Ganapol, B.D., and Furfaro, R. (2008) The Art of Analy- tiple sphere T-matrix Fortran code for use on parallel
tical Benchmarking (Lecture Notes in Computational computer clusters. Journal of Quantitative Spectros-
Science and Engineering, Vol. 62), Springer-Verlag, copy and Radiative Transfer, 112(13), 2182–2192.
New York, pp. 105–134. Maisch, M. (2000) The longterm signal of climate change
Gege, P. (1998) Characterization of the phytoplankton in in the Swiss Alps: Glacier retreat since the end of the
Lake Constance for classification by remote sensing. Little Ice Age and future ice decay scenarios. Geografia
Arch. Hydrobiolog. Adv. Limnology, 53, 179–193. Fisica e Dinamica Quaternaria, 23(2), 139–151.
Haeberli, W., and Beniston, M. (1998) Climate change Marston, P.L., Langley, D.S., and Kingsbury, D.L.
and its impacts on glaciers and permafrost in the Alps. (1982) Light scattering by bubbles in liquids: Mie
Ambio, 27(4), 258–265. theory, physical optical approximations and experi-
Hansen, J.E. and Travis, L.D. (1974) Light scattering in ments. Applied Science Research, 38, 373–383.
planetary atmospheres. Space Sci. Rev., 16, 527–610. Matzler, C. (2004) Matlab codes for Mie Scattering
Hapke, B.W. (1981) Bidirectional reflectance spectros- and absorption. Available at http://diogenes.iwt.
copy, 1: Theory, Journal of Geophysical Research, 86, unibremen.de /vt /laser /wriedt / Mie_Type_Codes / body_
3039–3054. mie_type_codes.html
Hapke, B.W. (1986) Bidirectional reflectance spectros- Meier, M.F., and Wahr, J.M. (2002) Sea level is rising:
copy, 4: The extinction coefficient and the opposition Do we know why? Proceedings of the National Acad-
effect. Icarus, 67, 264–280. emy of Science, 99(10), 6524–6526.
Hapke, B.W. (1993) Theory of Reflectance and Emittance Mishchenko, M.M., Dlugach, J.M., Yanovitskij, E.G.,
Spectroscopy, Cambridge University Press, New York. and Zakharova, E.T. (1999) Bidirectional reflectance
Hapke, B. (2002) Bidirectional reflectance spectroscopy, of flat, optically thick particulate layers: An efficient
5: The coherent backscatter opposition effect and radiative transfer solution and applications to snow
anisotropic scattering. Icarus, 157, 523–534. and soil surfaces. Journal of Quantitative Spectroscopy
Hapke, B. (2008) Bidirectional reflectance spectroscopy, and Radiative Transfer, 63, 409–432.
6: Effects of porosity. Icarus, 195, 918–926. Mobley, C.D. (1994) Light and Water Radiative Transfer
Heege, T. (2000) Flugzeuggestuzte Fernerkundung von in Natural Waters. Academic Press, New York.
Wasserinhaltsstoffen im Bodensee, PhD thesis, Remote Mobley, C.D., Sundman, L.K., and Boss, E. (2002) Phase
Sensing Technology Institute, German Aerospace function effects on oceanic light fields. Applied Optics,
Center (DLR). 41, 1035–1050.
Helfenstein, P., Veverka, J., and Thomas, P. (1988) Morris, R.D., Kottas, A., Furfaro, R., Taddy, M., and
Uranus satellites: Hapke parameters from Voyager Ganapol, B. (2008) A statistical framework for the
disk-integrated photometry. Icarus, 74, 231–239. sensitivity analysis of radiative transfer models used
Jin, Z., Charlock, T.P., Rutledge, K., Stamnes, K., and in remote sensed data product generation. IEEE Trans.
Wang, Y. (2006) An analytical solution of radiative Geosci. Remote Sens., 46(12), 4062–4074.
References 73

Mugnai, A., and Wiscombe, W.J. (1980) Scattering of the specific absorption of phytoplankton pigments,
radiation by moderately nonspherical particles. Jour- dissolved organic matter, and other particulate
nal of Atmospheric Science, 37, 1291–1307 materials. Limnol. Oceanogr., 26, 671–689.
Mullen, P.C., and Warren, S.G. (1988) Theory of the Seltzer, G.O. (1993) Late-Quaternary glaciation as a
optical properties of lake ice. Journal of Geophysical proxy for climate change in the central Andes. Moun-
Research, 93, 8403–8414. tain Research and Development, 13, 129–138.
Painter, T.H., Dozier, J., Roberts, D.A., Davis, R.E., and Shroder, J.F., Jr., and Bishop, M.P. (2000) Unroofing of
Green, R.O. (2003) Retrieval of subpixel snow- the Nanga Parbat Himalaya. In: M.A. Khan, P.J.
covered area and grain size from imaging spectrometer Treloar, M.P. Searle, and M.Q. Jan (Eds.), Tectonics
data. Remote Sensing of Environment, 85(1), 64–77. of the Nanga Parbat Syntaxis and the Western Hima-
Petzold, T.J. (1972) Volume Scattering Functions for laya (Special Publication No. 170), Geological Society,
Selected Ocean Waters (SIO Ref. 71–78), Scripps Insti- London, pp. 163–179.
tute of Oceanography, San Diego, CA. Siewert, C.E. (1978) The FN method for solving radia-
Piatek, J., Hapke, B., Nelson, R., Smythe, W., and Hale, tive-transfer problems in plane geometry. Astrophysics
A. (2004) Scattering properties of planetary regolith and Space Science, 58, 131–137.
analogs. Icarus, 171, 531–545. Siewert, C.E. (2000) A concise and accurate solution to
Picca, P. (2009) Applications of the Boltzmann equation Chandrasekhar’s basic problem in radiative transfer.
to the neutronics of nuclear reactors and to radiative Journal of Quantitative Spectroscopy and Radiative
transfer, PhD thesis, Politecnico di Torino. Transfer, 64, 109–130.
Picca, P., Furfaro, R., and Ganapol, B.D. (2007) Stamnes, K., Tsay, S.-C., Wiscombe, W.J., and Jaya-
Extrapolated iterative solution for the transport equa- weera, K. (1988) Numerically stable algorithm for dis-
tion in inhomogeneous media. Transactions of Amer- crete-ordinate-method radiative transfer in multiple
ican Nuclear Society, 97, 627–629. scattering and emitting layered media. Applied Optics,
Picca, P., Furfaro, R., and Ganapol, B.D. (2008a) 27, 2502–2509.
Optimization of the extrapolated iterative method for Warell, J., Sprague, A., Kozlowski, R., and Helbert, J.
the multislab transport problem. Paper presented at (2009) Surface composition and chemistry of Mercury:
PHYSOR 2008, Interlaken, September 14–19, 2008. Hapke modeling of MESSENGER/MASCS reflec-
Picca, P., Furfaro, R., Kargel, J., and Ganapol, B. tance spectra. Lunar Planet. Sci., 40, 1902.
(2008b) Forward and inverse models for photon trans- Warren, S.G. (1984) Optical constants of ice from the
port in soil–ice mixtures and their application to the ultraviolet to the microwave. Appl. Opt., 23(8/15).
problem of retrieving optical properties of planetary Warren, S.G., and Brandt, R.E. (2008) Optical constants
surfaces. Space Exploration Technologies (Proc. SPIE, of ice from the ultraviolet to the microwave: A revised
6960), Society of Photo-Optical Instrumentation compilation. Journal of Geophysical Research, 113,
Engineers, Bellingham, WA (11 pp.). D14220.
Pope, R.M., and Fry, E.S. (1997) Absorption spectrum Waterman, P.C. (1971) Symmetry, unitarity, and geom-
(380–700 nm) of pure water, II: Integrating cavity etry in electromagnetic scattering. Phys. Rev. D, 3, 825.
measurements. Appl. Opt., 36, 8710–8723. Wiscombe, W.J. (1980) Improved Mie scattering algo-
Previti, A. (2010) Semi-analytic radiative transfer model rithms. Applied Optics, 19(9), 1505–1509.
for remote sensing of planetary surfaces, Tesi Di Wiscombe, W.J. (1996) Mie Scattering Calculations:
Laurea Specialistica, Alma Mater Studiorum, Univer- Advances in Technique and Fast, Vector-Speed Com-
sità di Bologna. puter Codes (NCAR/TN-140+STR), National Center
Previti, A., Furfaro, R., Picca, P., Ganapol, B.D., and for Atmospheric Research, Boulder, CO.
Mostacci, D. (2011) Solving radiative transfer prob- Wiscombe, W. and Warren, S. (1980) A model for the
lems in highly heterogeneous media via domain decom- spectral albedo of snow, I: Pure snow. Journal of Atmo-
position and convergence acceleration techniques. spheric Science, 37, 2712–2733.
Applied Radiation and Isotopes, 69(8), 1146–1150. Yang, P., and Liou, K.N. (1998) Single-scattering
Prieur, L., and Sathyendranath, S. (1981) An optical properties of complex ice crystals in terrestrial
classification of coastal and oceanic waters based on atmosphere. Contrib. Atmos. Phys., 71, 223–248.